Radargrammetry with Chandrayaan-1 and Lro Mini-Rf Images of the Moon: Controlled Mosaics and Digital Topographic Models

41st Lunar and Planetary Science Conference (2010) 2428.pdf

RADARGRAMMETRY WITH CHANDRAYAAN-1 AND LRO MINI-RF IMAGES OF THE MOON: CONTROLLED MOSAICS AND DIGITAL TOPOGRAPHIC MODELS. R.L. Kirk1, D. Cook1, E. Howing- ton-Kraus1, J.M. Barrett1, T.L. Becker1, C.D. Neish2, B.J. Thomson2, D.B.J. Bussey2 and the Mini-RF Science Team. 1Astrogeology Science Center, U.S. Geological Survey, 2255 N. Gemini Dr., Flagstaff AZ 86001 ([email protected]), 2The Johns Hopkins University Applied Physics Laboratory, Laurel, MD 20723.

Introduction: The Mini-RF investigation [1] consists of takes account of the different principles by which a radar two synthetic aperture radar (SAR) imagers for lunar remote image is formed. The fundamental tool in both cases is a sensing: Mini-SAR (also known as “Forerunner”), which sensor model, which allows one to calculate the image coor- was flown on the ISRO Chandrayaan-1 orbiter in November dinates (line and sample) of any point whose latitude, longi- 2008, and the Mini-RF technology demonstration, which was tude, and elevation are specified, or the latitude and longi- flown on the NASA Lunar Reconnaissance Orbiter (LRO) in tude of any image pixel provided the elevation is specified. June 2009. A primary objective for both instruments was to The sensor model is needed to transform image pixels into investigate the possible presence of water ice in permanently their appropriate locations in map coordinates. It is also used shadowed areas near the lunar poles [2]. Not only can these to make controlled image mosaics that have improved accu- imagers “see in the dark” by providing their own illumina- racy by bundle adjustment, in which the spacecraft positions tion, they measure the full polarization characteristics of the are adjusted to achieve the best least-squares agreement be- reflected signal for circularly polarized transmitted radiation tween the positions of features in overlapping images and [3]. A strong return in the “unexpected” (i.e., not reversed as between the images and pre-existing ground control. Finally, it would be if reflected from a smooth interface) sense of the sensor model is used to turn a dense set of feature corre- circular polarization can be indicative of radar-transparent spondences between a pair of images, obtained either by materials such as ice (e.g., [4]). Mini-SAR obtained nearly automated image matching or interactively, into a DTM. complete image coverage of both lunar poles to 80° latitude Our approach to radargrammetric processing of Mini-RF with a resolution of 150 m and radar wavelength of 12.6 cm images follows that which we have applied to numerous (S Band), as well as images of non-polar targets for compari- optical sensors and to the Magellan and Cassini radar im- son purposes. LRO Mini-RF is capable of imaging in both agers [10–12]. In particular, we use the USGS in-house car- S-Band and X-band (3 cm) wavelengths and at 150 m and 30 tographic software system ISIS [13] to ingest and prepare the m (zoom mode) resolutions. (The 3-cm wavelength is actu- data, project images onto a known reference surface (sphere, ally in the C-Band but is referred to as X-Band because early ellipsoid, or DTM), and perform a variety of general image designs used a wavelength in this band.) Most of its observa- analysis and enhancement tasks. We use a commercial digi- tions to date have been obtained in S-Band zoom mode, and tal photogrammetric workstation running SOCET SET (® include both right- and left-looking coverage of much of the BAE Systems) software [14] for DTM production by auto- south polar zone in support of the LCROSS mission [5] as mated matching and for interactive editing of DTMs using its well as extensive non-polar coverage. stereo display capability. Using this commercial system with In this abstract we describe the software and techniques a given data set requires not only an appropriate sensor we are developing for radargrammetric analysis of the Mini- model, but also software to translate the images and support- RF images. These tools will enable us to make controlled ing information from ISIS to SOCET SET formats. image mosaics with both image-to-image seam errors and A key difference between Mini-RF and the Magellan and absolute positional accuracy improved by at least an order of Cassini SARs is that data from these earlier instruments were magnitude compared to uncontrolled products. They will produced only in map-projected form. As a result, we used also permit the derivation of digital topographic models ISIS capabilities for transforming data from one projection to (DTMs) from stereo pairs of radar images, from which ortho- another to make uncontrolled mosaics but did not implement images (images from which topographic parallax distortions an ISIS sensor model for these missions. Instead, we created have been removed), slope maps, and other products can be sensor models for SOCET SET that worked by first “undo- generated. A primary motivation for both types of processing” the map projection to get back to the fundamental radar ing is to coregister the radar images as closely as possible to image coordinates of range and Doppler shift. These sensor one another and to other datasets [6], so that detailed analy- models perform rather complex bookkeeping so they work ses can be made. Comparison of multiple polarization meas- with mosaics of multiple orbits (Magellan) or multiple radar urements, correction of these results for locally varying inci- beams (Cassini). They can be used to perform bundle ad- dence angles (i.e., slope effects), and improved topographic justments and produce DTMs. In contrast, the Mini-RF data data for determining what areas are permanently shadowed [15] are produced and archived both in “Level 2” map coor- are logical extensions of preliminary work such as [7] and dinates [16] and in “Level 1” coordinates that are similar but should all lead to firmer conclusions about the presence of not identical to the raw camera geometry for a pushbroom ice at the lunar poles. Precise georeferencing of the images scanning camera. will also be essential to using them in studies of the LCROSS ISIS Software: Given the availability of Mini-RF data impact point. Coregistration of Mini-RF images with a vari- in Level 1 instrument coordinates, we have implemented a ety of spectrophotometric data should also shed light on in- sensor model for the ISIS 3 system, based on the physical teresting sunlit features, for example pyroclastic deposits [8]. principals of SAR image formation. We have verified that We note in passing that Mini-RF, with intermediate reso- map-projecting Level 1 images with this sensor model yields lution between those of the LROC narrow-angle and wide- results consistent with the Level 2 products from the Mini- angle cameras, is ideally suited for completing a uniform RF processing pipeline developed by Vexcel Corporation. global topographic model of the Moon. Data from the This apparently redundant capability is valuable for several LOLA laser altimeter have unprecedented meter-scale accu- reasons. First, image projection in ISIS can potentially make racy, but LOLA tracks will typically be separated by a kilo- use of improved spacecraft trajectory data in the form of meter or more at low latitudes [9]. These gaps could be SPICE kernels [17] and improved topographic models onto filled by DTM data derived from zoom mode stereopairs and which to project the images as these become available. Sec- controlled to be consistent with LOLA by using the tech- ond, we have also incorporated the Mini-RF sensor model niques described here. into the ISIS bundle-adjustment program “jigsaw” so that Radargrammetry: Radargrammetry is the art and sci- images can be controlled to yield even higher accuracy. ence of making geometric measurements based on radar Finally, and most importantly, the ISIS software (unlike ei- images, and is precisely analogous to photogrammetry but ther SOCET SET or the Vexcel SAR processor) is freely 41st Lunar and Planetary Science Conference (2010) 2428.pdf

available and can be used by anyone to make controlled mo- and horizontal resolution of at least 500 m. We plan to ex- saics and other products from Mini-RF images. The system periment with DTM production from these image combina- includes software needed to ingest both Level 1 and Level 2 tions in order to verify the precision and resolution that can data products archived in NASA Planetary Data System be achieved in practice. (PDS) format. All software elements are equally compatible A total of 15 non-polar images were obtained by with both Chandrayaan-1 and LRO images. Chandrayaan-1, one of which has already been re-imaged by SOCET SET: We are currently implementing a Mini- LRO. As described above, radar stereo imaging by LRO has RF sensor model for SOCET SET, as well as the software tremendous potential for filling in the topographic map of the needed to translate images and spacecraft trajectory data (in Moon’s equatorial zone. DTM properties for these types of SPICE SPK format) to SOCET formats. The sensor model stereopairs would be similar to the LRO-Chandrayaan and operates on identical principles to that already developed for LRO-LRO polar data sets discussed above. ISIS, and is substantially simpler than our past Magellan [9] Concluson: We have implemented a series of radar- and Cassini [10] models. Once the model is completed, all grammetric tools for processing Chandrayaan-1 and LRO SOCET SET functions will be available for Chandrayaan-1 Mini-RF radar images of the Moon. These tools will be used and LRO Mini-RF images in combination with one another within the Mini-RF team to make high-precision controlled or, if desired, with any other supported images. The bundle mosaics and digital topographic models. Through the open adjustment capability of SOCET SET will provide a useful release of ISIS software and the USGS/NASA Planetary check of the validity of results from ISIS “jigsaw” but the Photogrammetry Guest Facility, the same capabilities will main application will be DTM production by a combination soon be available to the entire planetary research community. of automatic image matching and interactive editing. The References: [1] Nozette, S. et al. (2009) Space Sci. Rev., in commercial software and special display hardware it uses are press. [2] Bussey, D.B.J. et al. (2003) GRL, 30, 7158. [3] Raney, relatively expensive, but several planetary investigators have R.K. (2007) IEEE Trans Geosci. Remote Sens. 45, 3397. [4] SOCET workstations at their home institutions (see, e.g., Harmon, J.K. et al. (2001) Icarus, 149, 1. [5] Neish, C.D. et al. [18]). The USGS has several such workstations used for (2010) LPS XLI, this conference. [6] Archinal, B.A. et al. (2010) LPS mapping under the NASA Planetary Cartography program XLI, this conference. [7] Thomson, B.J. et al. (2010) LPS XLI, this and various flight programs, and one that is operated as a conference. [8] Carter, L.M., et al. (2010) LPS XLI, this conference; guest facility at which outside researchers can make their Trang, D. et al. (2010) LPS XLI, this conference. [9] Smith, D.E. et own DTMs from publically available data [19]. al. (2009) Space Sci. Rev. doi:10.1017/s11214-009-9512-y. [10] Test Data and Prospects: By far the most extensive Howington-Kraus, E. et al. (2002) LPS XXXIII, 1986. [11] Kirk, R.L. sets of overlapping SAR images are in the areas poleward of 80° north and south. Forerunner images covering >90% of et al. (2010) Icarus, in revision. [12] Kirk, R.L. et al. (2008) these regions were obtained. LRO Mini-RF S-zoom mode IAPRSSIS XXXVII(4), 973. [13] Anderson, J.A. et al. (2004) LPS images cover a smaller but still appreciable fraction of the XXXV, 2039. [14] Miller, S.B., and A.S. Walker (1993) south polar zone at higher resolution [5]. Both datasets pro- ACSM/ASPRS Ann. Conv. 3, 256; — (1995) Z. Phot. Fern. 63(1) 4. vide a mixture of eastward and westward viewing and illu- [15] Reid, M. (2009) SIS For Mini-RF Advanced Technologies Lu- mination direction; although the look direction (left or right nar Reconnaissance Orbiter (LRO) Payload Operations Center, of the ground track) is constant for each image strip, the Document MRF-4009. [16] Batson, R.M. (1990) in Planetary Car- compass directions of illumination on the ascending and tography, 60–95. [17] Acton, C.H. (1999) LPS XXX, 1233. [18] descending sides of the orbit are opposite. The south polar Beyer, R. et al. (2010) LPS XLI, this conference. [19] Kirk, R.L. et zone, in particular, the 98 km diameter crater Cabeus in al. (2009) LPS XL, 1414. which the LCROSS spacecraft impacted, is therefore the highest priority for controlled mosaic and DTM production, results of which will be presented in our poster. Visual in- spection indicates that image-to-image offsets of as much as several km are found in the image sets from both missions, despite the expectation of spacecraft positioning errors 500 m for Chandrayaan-1 and 100 meters for LRO. Our expe- rience with Cassini data indicates that radargrammetric bundle adjustment can reduce these relative errors to on the order of the pixel size (roughly half the image resolution) even in the presence of substantial speckle noise in the images. Thus, precisions on the order of 100 m and 10 m or better should be achievable for Chandrayaan-1 and LRO zoom data, re- spectively. The accuracy of absolute positioning depends on the availability of suitable reference data to provide ground control and is probably on the order of 100 m for LOLA altimetric data in their initial state. The properties of stereo DTMs depend on the resolution and viewing directions of the source imagery. The opposite- look LRO S-zoom images (Fig. 1) have high resolution and strong image convergence. Standard formulas of radargrammetry indicate that the expected vertical precision (EP) for these images is on the order of 10 m, if image matching is precise to half the resolution as we have found to be the case for Cassini radar [11]. Horizontal resolution of the DTM can Figure 1. Anaglyph formed by combining uncontrolled mosaics of be expected to be a few times the resolution, or on the order of 50-100 m. The differences in surface shading may, how- left- and right-looking LRO Mini-RF S-zoom mode images, to be ever, make matching of these oppositely illuminated images viewed with red lens on the left eye and blue-green on the right. challenging. Same-side images from LRO and Chandrayaan- Polar Stereographic projection with south approximately at the bot- 1, which have incidence angles of 47.6° and 33.5° respec- tom. The 98-km crater Cabeus occupies the bottom portion of the tively, could be used but the relatively weak geometry and area of overlapping coverage. Cabeus B is to the upper left and worse resolution of the latter would lead to an EP of ~120 m Cabeus A at top center.